feat(registry): add 9 high-value manipulation datasets + access filter

[arpitg1304]

Summary

Grows the dataset registry 23 → 32 with widely-used, well-documented, openly-accessible manipulation datasets, chosen to close the biggest coverage gaps (sim data-generation, bimanual/humanoid, and the human-collection paradigm). Also makes access restrictions first-class so the registry stays a trustworthy "here's what you can actually use" catalog.

New datasets

Tier 1 — must-add

id	why	license	format
robocasa	standard sim-kitchen corpus behind GR00T	MIT	hdf5
maniskill	GPU-parallel sim benchmark, millions of frames	Apache-2.0	hdf5
calvin	reference long-horizon language-conditioned benchmark	MIT	other
robomimic	canonical IL / offline-RL benchmark	MIT	hdf5
robomind	multi-embodiment incl. Tien Kung humanoid (🔒 registration)	Apache-2.0	hdf5

Tier 2 — strong additions

id	why	license	format
agibot_world_beta	full-scale successor to Alpha, 1M+ traj / 2,976h (⊘ non-commercial)	CC-BY-NC-SA-4.0	lerobot-v2
robotwin	bimanual data generator, 100k+ traj, 5 embodiments	Apache-2.0	hdf5
mimicgen	auto-generated demos, robomimic-format	CC-BY-4.0	hdf5
umi	in-the-wild human handheld-gripper collection	MIT	zarr

Access restrictions are now first-class

• New tag-vocabulary entries registration_required and non_commercial (documented in CONTRIBUTING.md).
• Registry HTML gains a dedicated Access filter (Open / Registration required / Non-commercial) and a rose gated badge on restricted cards; access tags are excluded from the general Tags dropdown to avoid redundancy.
• robomind → 🔒 registration; agibot_world_beta → ⊘ non-commercial.

Backlog

Tier-3 / deferred candidates (DexMimicGen, EgoDex, FMB, BEHAVIOR-1K, Mobile ALOHA, RoboCasa365, ARIO) are captured in forge/registry/CANDIDATES.md for later promotion.

Validation

• forge registry validate → 0 errors (2 pre-existing warnings, unrelated).
• All 9 entries construct cleanly through DatasetEntry.from_dict.
• docs/registry.html regenerated (32 cards); rendered and visually verified — Access filter present, gated badges display correctly.
• Registry test suite passes (the one failing TestFormatRegistry::test_detect_format_zarr is a pre-existing zarr-API drift in a test fixture, unrelated to this change).

Note: all new entries are forge_verified: false — source URIs/paths still need maintainer verification (forge registry validate --probe). added_at is 2026-07-21. agibot_world_beta is intended to eventually supersede agibot_world_alpha.

🤖 Generated with Claude Code